Why the brands winning in AI-powered search are the ones who never stopped doing SEO properly and why the shortcuts are already backfiring.
There’s a version of the current AI search moment that gets talked about a lot in marketing circles. It goes something like this: Google has changed, search has changed, AI is now answering questions directly, so the old rules don’t apply and you need a new strategy built around AI visibility, GEO, AEO, citations, and mentions.
We understand why that narrative is appealing. It’s new, it’s urgent, and it gives people something concrete to sell.
It’s also, in most of the ways that matter, wrong.
What’s actually happening in AI search is more interesting and more demanding than the “everything has changed” story suggests. And if you’re a business trying to navigate it, understanding the difference between the hype and the reality could be the thing that protects your organic performance over the next 12 to 18 months.
What Google actually said about AI search and SEO
Google recently published official guidance on how to optimise for its generative AI features, including AI Overviews and AI Mode. For anyone expecting a new set of rules, the guidance was something of a cold shower. The message, stated plainly, was this: the best practices for SEO continue to be relevant, because our generative AI features are rooted in our core search ranking and quality systems.
In other words, AI Overviews and AI Mode are not separate systems running on separate signals. They’re built on top of the same index, the same quality assessments, and the same ranking systems that determine standard organic search results. The AI layer retrieves content from pages that already rank well, applies retrieval-augmented generation (RAG) to pull specific information from those pages, and generates responses grounded in that content.
What this means practically is that the path to appearing in AI search responses runs directly through doing good SEO. Crawlable, indexable content. Strong technical foundations. Authoritative, original pages that demonstrate genuine expertise. Credible third-party signals. Everything that has always mattered in organic search still matters because the AI is reading from the same source.
Google was also explicit about what doesn’t matter for AI search. You don’t need to create special AI text files like llms.txt. You don’t need to “chunk” your content into fragments. You don’t need to rewrite everything in a specific style for AI systems. You don’t need to pursue inauthentic brand mentions. These tactics, Google confirmed, have no special effect on AI search visibility.
What the data is already showing about AI content shortcuts
While Google was publishing its guidance, SEO consultant Lily Ray was publishing something that every marketing team evaluating AI content tools should read carefully.
Over several months, she tracked more than 220 websites that had been publicly identified, either by themselves or by their AI content vendors, as users of AI content creation and scaling platforms. She wanted to know what happened after the case study headlines. The pattern that emerged across those 220+ sites was consistent and stark. 54% of the sites she monitored lost 30% or more of their peak organic traffic. 39% lost 50% or more. 22% lost 75% or more. In many cases, the eventual traffic loss exceeded the peak gain, meaning the sites ended up worse off than before they started scaling AI content.
The trajectory was remarkably similar across industries: a rapid growth in organic pages over six to twelve months, a traffic peak three to six months after the content peak, and then a steep decline that typically erased most of the gain within the following year. Glenn Gabe has called it “Mount AI” steep growth, followed by an equally steep drop, once Google’s systems have gathered enough signals to understand what is happening.
Most of the sites that declined were using some combination of eight content patterns that Ray identified as high-risk: comparison pages at scale, “What is X” glossary pages designed for AI citation, “Best X for Y” listicles, self-promotional listicles in which the publisher names itself the top-ranked option, competitor alternative pages, programmatic location and language pages, FAQ farms, and off-topic content published at volume.
The common thread across all eight is that they are templates designed to influence rankings and AI citations, rather than content created because a real user genuinely needs it. They are detectable as a pattern. And when enough sites implement the same pattern at scale, Google’s systems become very good at identifying and demoting it.
Ray also identified a likely unconfirmed Google update in late January 2026, after which at least 40 sites she was monitoring saw organic traffic declines of between 40% and 95% most of them concentrated in blog subfolders where self-promotional listicles and other AI-generated content had been published at volume. Some of those sites saw the impact spread from the subfolder to the full domain.
Why this matters for AI search, not just traditional SEO
Here’s where the two stories connect, and where the stakes become clear.
AI search experiences, AI Overviews, AI Mode, the generative responses being built into search engines are powered by RAG. They retrieve content from Google’s index and use it to generate responses. What gets retrieved is what ranks. What ranks is determined by Google’s quality systems. And Google’s quality systems are specifically tuned to detect and demote the same kinds of low-quality, template-driven, scaled content that Ray’s data shows is already collapsing in traditional search.
This means that the shortcut being sold to many businesses right now “scale AI content to win AI citations” is exactly backwards. The sites producing templated, formulaic content at scale to capture AI mentions are doing the thing most likely to get them demoted from the index that AI search reads from.
Put differently: bad SEO is bad GEO. The signals that cause a site to lose visibility in traditional search are the same signals that cause it to lose visibility in AI-generated responses. They are not separate problems with separate solutions.
The brands that are winning in AI search
Ray’s analysis included an observation that didn’t get as much attention as the decline data, but is arguably the more important finding: the brands still growing across her dataset were, broadly, the ones whose content did not match the eight risky templates. This is not a coincidence. It reflects something that Google has been consistent about for years and that its new AI search guidance restates clearly: the signal that a page is worth surfacing is whether real users would find it genuinely helpful, original, and trustworthy. That signal doesn’t change because the delivery mechanism has moved from a blue link to an AI-generated paragraph.
The brands winning in AI search tend to share certain characteristics. Their content demonstrates real expertise, the kind that comes from people who actually know the subject, not from a prompt that summarises what’s already on the first page of results. Their pages contain original information: first-hand experience, proprietary data, unique perspectives, specific examples. Their technical foundations are solid enough that Google can find and index everything they publish. And their authority is supported by genuine third-party signals; coverage, citations, and links earned because they’ve said or done something worth talking about.
These are SEO fundamentals. They have always been SEO fundamentals. The AI layer has raised the stakes for doing them well, because the content that surfaces in AI responses needs to be both highly relevant and demonstrably trustworthy and because the gap between sites that do this and sites that are trying to shortcut it is becoming increasingly visible in the data.
What this means for how you should be thinking about AI content tools
We want to be clear: AI content tools are not the problem, the problem is how they’re being used.
There are genuinely valuable applications for AI in a content workflow. Research and synthesis. Brief creation. Pulling together proprietary data and presenting it clearly. Identifying content gaps. Supporting writers who are subject matter experts but not natural writers. Accelerating the production of content that still has expert oversight, fact-checking, and editorial review at every stage.
What is risky and what the data increasingly confirms is risky, is using AI to publish pages at volume without those quality controls in place. When the goal becomes the number of pages rather than the usefulness of each one, the content that gets produced tends to look like the templates Ray identified. It gets ranked initially because it’s relevant. It loses those rankings when Google’s systems gather enough signals to understand that it isn’t genuinely useful. Ray’s diagnostic questions are worth keeping on hand whenever you’re evaluating a content programme:
- Does this page exist because a real user genuinely needs it, or because a search engine or LLM might cite it?
- Could a competitor produce a near-identical version of it tomorrow using the same prompt?
- Is there anything on this page; first-hand experience, proprietary data, a genuine point of view, that isn’t already available in the top ten results for this query?
If the honest answer to that last question is no, the page is probably not worth publishing.
The practical upshot
AI search is real, and the changes happening in how people discover, compare, and evaluate brands are significant. We’re not suggesting otherwise.
But the response to those changes should not be to separate AI search optimisation from SEO quality and treat them as different disciplines requiring different approaches. That separation is exactly the error that’s producing the “Mount AI” trajectories Ray is documenting.
The right response is to treat AI search as a reason to do SEO better, not differently. That means strengthening the technical foundations so that everything is crawlable and indexable. It means improving content usefulness and originality so that what you publish actually contains something a person couldn’t find five seconds ago on a competitor’s site. It means earning credible third-party signals through coverage, links, and brand mentions that are the product of genuine reputation, not manufactured footprints. And it means being deliberate about what AI systems can learn and infer about your brand from everything that exists about you on the open web.
The brands that will win in AI search, the ones that will be cited, recommended, and trusted by AI systems over the next few years will be the ones producing content that is genuinely worth citing.
That’s always been the situation. The stakes are just higher now.
If you’re reviewing your content strategy in light of these changes and want a clear-eyed view of where your risks and opportunities lie, we’d be happy to take a look. Get in touch with the Koozai team.
Leave a Reply